Correspondence

Correspondence

IN REPLY: We agree with Dr Atkin’s description of the conclusions from the report of the International Workshop on Screening for Breast Cancer. 1 The statement reporting a mortality benefit described for women older than age 40 years is supported by a meta-analysis that included several randomized trials and specifically evaluated benefits for women age 40 to 49 years.2 We regret that this additional reference was not attached to this statement. We agree with Dr Atkins that there remains uncertainty regarding long-term radiation exposure from mammography, particularly in high-risk women who begin screening earlier than age 40 years. As we hoped to convey, we feel that alternative imaging modalities, such as breast magnetic resonance imaging, which do not involve radiation exposure may ultimately replace mammography in individualized programs for breast cancer prevention.3

Angela R. Bradbury and Olufunmilyo I. Olopade Department of Medicine, Section of Hematology/Oncology, University of Chicago, Chicago, IL

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The authors indicated no potential conflicts of interest. REFERENCES 1. Fletcher SW, Vlack W, Harris R, et al: Report of the International Workshop on Screening for Breast Cancer. J Natl Cancer Inst 85:1644-1656, 1993 2. Hendrick RE, Smith RA, Rutledge JH III, et al: Benefit of screening mammography in women aged 40-49: A new meta-analysis of randomized controlled trials. J Natl Cancer Inst Monogr 22:87-92, 1997 3. Bradbury A, Olopade OI: The case for individualized screening recommendations for breast cancer. J Clin Oncol 24:3328-3329, 2006

■ ■ ■

Should Subgroup Analysis of Randomized Clinical Trials Have a Direct Impact on Clinical Practice? TO THE EDITOR: The recent publication in the Journal of Clinical Oncology of the randomized clinical trial performed by the European Organisation for Research and Treatment of Cancer Chronotherapy Group did raise the issue of the clinical reliability of sex subgroup analysis in clinical trials.1 The primary end point of this large cooperative multicentric phase III trial was 2-year survival, and the sample size was then calculated to determine a 10% survival difference (from 30% to 40%) in favor of the chronomodulated arm (superiority trial). Patients were stratified according to performance status, liver involvement, and center. In the abstract (which has to be considered the most rapid source for a take-home message), the authors conclude that “both regimens achieved similar median survival times more than 18 months with an acceptable toxicity” and that “chronomodulated schedule produced a survival advantage in men.”1 A controversy exists regarding the correct interpretation of subgroup analysis of randomized clinical trials, given that the risk to provide misinterpretation could lead to wrong treatment guidelines. Although the trial is adequately powered for its main end point, it is not able to detect differences in efficacy between subgroups with adequate power. Moreover, the sex analysis lead to both a significantly better effect for chronotherapy in men, but also a significantly better effect for standard chemotherapy in women, and this is not mentioned in the abstract conclusion.1 Actually, in the article a balanced discussion about the subgroup issues is present. Although the multivariate analysis should protect over baseline prognostic factors’ interaction (when adequately studied), the list of the patients’ characteristics, which the authors decided to consider for subgroup analysis, should be reported in the article as well. A number of pieces of specific literature have been published regarding the risk of the erroneous interpretation of subgroup analyses and the most common conclusions are those analyses can be overinterpreted and may lead to suboptimal patient care.2 A sort of methodologic recommendation for subgroup analyses and their interpretation has been recently produced by Brookes et al3,4— only previously planned subgroup analyses should be performed; any trial should be powered taking into account the prespecified subgroup analyses; notwithstanding all these considerations, any specific analy-

sis is affected by several bias, and the conclusions need to be softened. In absence of strong common evidence, this is an hypothesisgenerating exercise for further specifically addressed randomized trial. The risk in subgroup analysis to provide errors has been also quantified, and seems to be not related to the patient sample size.2 Correctly, Giacchetti et al1 performed a sex-treatment interaction analysis, and this interaction resulted significantly at the multivariate analysis. This result is interesting. However, the observed survival differences by sex could be due to imbalance in unmeasured baseline prognostic factors (demographic and/or molecular characteristics). According to article’s results, “Patients’ characteristics according to sex were well balanced except for age and PS. Age ⱕ 50 years and PS of 1 or 2 were more frequent in women than in men (21% v 14%; and 59% v 46.5%, respectively). After stratification for age and PS, the interaction tests were not found significant (P ⬎ .1).” If not due to imbalance, the difference of outcome between males and females could be a true difference or could be due to statistical chance alone, especially because not only sex but a long list of patient characteristics has been analyzed. So, in our opinion, more caution should be used when authors decide to put the results of secondary, exploratory analyses in the abstract of an article.

Emilio Bria Department of Medical Oncology, Regina Elena National Cancer Institute, Rome, Italy

Massimo Di Maio National Cancer Institute, Naples, Italy

Federica Cuppone, Cecilia Nistico`, and Francesco Cognetti Department of Medical Oncology, Regina Elena National Cancer Institute, Rome, Italy

Diana Giannarelli Biostatistics, Regina Elena National Cancer Institute, Rome, Italy

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The authors indicated no potential conflicts of interest. REFERENCES 1. Giacchetti S, Bjarnason G, Garufi C, et al: Phase III trial comparing 4-day chronomodulated therapy versus 2-day conventional delivery of fluorouracil, leucovorin, and oxaliplatin as first-line chemotherapy of metastatic colorectal cancer: The European Organisation for Research and Treatment of Cancer Chronotherapy Group. J Clin Oncol 24:3562-3569, 2006 2. Lagakos SW: The challenge of subgroup analyses–reporting without distorting. N Engl J Med 354:1667-1669, 2006 605

www.jco.org

Downloaded from jco.ascopubs.org on November 24, 2009 . For personal use only. No other uses without permission. Copyright © 2007 by the American Society of Clinical Oncology. All rights reserved.

Correspondence

3. Brookes ST, Whitley E, Peters TJ, et al: Subgroup analyses in randomized controlled trials: Quantifying the risks of false-positives and false-negatives. Health Technol Assess 5:1-56, 2001

4. Brookes ST, Whitely E, Egger M, et al: Subgroup analyses in randomized trials: Risks of subgroup-specific analyses; power and sample size for the interaction test. J Clin Epidemiol 57:229-236, 2004

■ ■ ■

IN REPLY: In their Conclusion, Bria et al emphasize that more caution is needed in reporting the effects of sex on survival and tolerability that were derived from the European Organisation for Research and Treatment of Cancer phase III trial 05963.1 Their comment is based on the controversy regarding the generation and the interpretation of subgroups analyses. Indeed, we strongly believe that the abstract serves the purpose of delivering the most clinically relevant findings with honesty. This is why we wrote, “In women, the risk of an earlier death on ChronoFLO4 was increased by 38% compared with FOLFOX2. . . .In men, the risk of death was decreased by 25% on ChronoFLO4. . . .” So, we agree with Bria et al that the abstract is the most rapid source for a take home message, but we mean the whole abstract, not the conclusion only. For treatment interaction tests, we considered factors related to patient characteristics (age, sex, WHO performance status [PS] 0 v ⬎ 0), baseline hematologic and biochemical data (leukocyte, granulocyte and platelet counts, alkaline phosphatases, gamma glutamyltransferase, transaminases, carcinoembryonic antigen, and CA 19-9), and tumor characteristics (primary site, Duke’s stage, liver involvement, number of metastatic sites). At the exception of sex, no other factor displayed any strongly significant interaction with the delivery schedule (P ⬍ .01). As mentioned in the article, an imbalance between women and men was found for PS and age. After stratification for age and PS, the interaction tests were not found significant (P ⬎ .1) because of a lack of power, but presented the same trend for a higher efficacy in men treated by infusion of fluorouracil, leucovorin, and oxaliplatin for 4 days (chronomodulated FLO4) in the different age and PS strata. Indeed, the difference between men and women could be a true difference or statistical chance. However, in a Cox proportional hazards model a sex x schedule interaction term was found significant in presence of PS, age, number of metastatic sites, and percentage of liver involvement (Table 4).1 This supports that sex effect is a true difference. Lagakos recommends being cautious with subgroups analysis. He advises to perform planned subgroups analysis, appropriate analysis with interaction tests, and to give magnitude of the treatment difference with corresponding confidence interval instead of the P value.2 Bria et al acknowledge that we followed these recommendations, including interaction test and Forrest plot of interaction between schedule and sex. However, we could not plan the subgroups analyses since such effect was largely unknown when we initiated this trial. Thus, a consistent impact of sex on chemotherapy toxicity or patient outcome started to arise in the literature once accrual to our trial was completed.3 We feel that clinical research should aim at discovering new prognostic and predictive factors, new treatments, better treatment schedules, and new strategies. Through the use of chronomodulated delivery of chemotherapy, our group first demonstrated the activity of oxaliplatin against colorectal cancer,4 then the relevance of liver metastases surgery for the survival of patients with downstaged previously unresectable metastases.5,6 Both of these findings initially encountered great scepticism, yet they now have profoundly modified the current treatment of colorectal cancer worldwide. Thus we share the opinion of Lagakos that avoiding any presentation of subgroup analysis be606

cause of their history of being overinterpreted would have been a steep price to pay. Rather than advocating for a ready change in clinical practice which could imply the delivery of FOLFOX (2-hour infusion of oxaliplatin and leucovorin on day 1 and 2-hour infusion of leucovorin on day 2, which intercalculated 22-hour constant infusion rate of fluorouracil on days 1 and 2) in women and that of chronomodulated FLO4 in men as standard best practice, we proposed to investigate the mechanisms through which optimal scheduling of chemotherapy differs between men and women. Indeed, increased attention is currently being paid by the medical and scientific community to the role of sex for treatment toxicities and patient outcome in cancer and other diseases. Our correspondence (Bria et al and our reply) emphasizes the need for studies testing new chemotherapy and chronotherapy associations with targeted molecules and a priori planning of effects of sex or other factors identified in subgroup analyses. Already a sex dependency of toxicity and survival has been recently found by our group for two separate studies, including one with cetuximab.7,8 We are currently developing new protocols aimed at the determination of the best schedule in women with metastatic colorectal cancers. Moreover, to better understand the mechanism of the sex effect, translational research projects of molecular clock determinants of optimal schedule are also ongoing. The challenge ahead of us is the administration of tailored chronotherapy (ie, the determination of the right drugs at the right doses and at the right time in the right patient).

Sylvie Giacchetti INSERM U 776 and Hôpital Paul Brousse, Assistance Publique-Hôpitaux de Paris, Paris, France

Thierry Gorlia Data Center, European Organisation for Research and Treatment of Cancer, Brussels, Belgium

Carlo Garufi Istituto Regina Elena, Roma, Italy

Francis Le´vi INSERM U 776, University Paris Sud, and Hôpital Paul Brousse, Assistance Publique-Hôpitaux de Paris, Paris, France

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The authors indicated no potential conflicts of interest. REFERENCES 1. Giacchetti S, Bjarnason G, Garufi C, et al: Phase III trial comparing 4-day chronomodulated therapy versus 2-day conventional delivery of fluorouracil, leucovorin, and oxaliplatin as first-line chemotherapy of metastatic colorectal cancer: The European Organisation for Research and Treatment of Cancer Chronotherapy Group. J Clin Oncol 24:3562-3569, 2006 2. Lagakos SW: The challenge of subgroup analyses–reporting without distorting. N Engl J Med 354:1667-1669, 2006 3. Neugut AI, Jacobson JS: Women and lung cancer: Gender equality at a crossroad? JAMA 296:218-219, 2006 4. Lévi F, Misset JL, Brienza S, et al: A chronopharmacologic phase II clinical trial with 5-fluorouracil, folinic acid and oxaliplatinum using an ambulatory multichannel programmable pump: High antitumor effectiveness against metastatic colorectal cancer. Cancer 69:893-900, 1992 JOURNAL OF CLINICAL ONCOLOGY

Downloaded from jco.ascopubs.org on November 24, 2009 . For personal use only. No other uses without permission. Copyright © 2007 by the American Society of Clinical Oncology. All rights reserved.